X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/131fcb74c972adcf3455357f4ec7df0224c08f37..ea74f5d95928a521a588737e81f1de94eef25d19:/src/kernel/activity/ExecImpl.cpp diff --git a/src/kernel/activity/ExecImpl.cpp b/src/kernel/activity/ExecImpl.cpp index 166d78f3d1..9b97aab59e 100644 --- a/src/kernel/activity/ExecImpl.cpp +++ b/src/kernel/activity/ExecImpl.cpp @@ -1,56 +1,21 @@ -/* Copyright (c) 2007-2021. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2022. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ -#include "simgrid/s4u/Exec.hpp" +#include +#include +#include +#include + #include "src/kernel/activity/ExecImpl.hpp" -#include "simgrid/Exception.hpp" -#include "simgrid/modelchecker.h" +#include "src/kernel/actor/ActorImpl.hpp" +#include "src/kernel/actor/SimcallObserver.hpp" +#include "src/kernel/resource/CpuImpl.hpp" #include "src/mc/mc_replay.hpp" #include "src/surf/HostImpl.hpp" -#include "src/surf/cpu_interface.hpp" -#include "src/surf/surf_interface.hpp" - -#include "simgrid/s4u/Host.hpp" - -#include -XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process); - -void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[], - size_t count, double timeout) -{ - if (timeout < 0.0) { - simcall->timeout_cb_ = nullptr; - } else { - simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() { - simcall->timeout_cb_ = nullptr; - for (size_t i = 0; i < count; i++) { - // Remove the first occurrence of simcall: - auto* exec = execs[i]; - auto j = boost::range::find(exec->simcalls_, simcall); - if (j != exec->simcalls_.end()) - exec->simcalls_.erase(j); - } - simcall_execution_waitany_for__set__result(simcall, -1); - simcall->issuer_->simcall_answer(); - }); - } - - for (size_t i = 0; i < count; i++) { - /* associate this simcall to the the synchro */ - auto* exec = execs[i]; - exec->simcalls_.push_back(simcall); - - /* see if the synchro is already finished */ - if (exec->state_ != simgrid::kernel::activity::State::WAITING && - exec->state_ != simgrid::kernel::activity::State::RUNNING) { - exec->finish(); - break; - } - } -} +XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(ker_actor); namespace simgrid { namespace kernel { @@ -58,10 +23,10 @@ namespace activity { ExecImpl::ExecImpl() { - piface_ = new s4u::Exec(this); + piface_ = new s4u::Exec(this); actor::ActorImpl* self = actor::ActorImpl::self(); if (self) { - actor_ = self; + set_actor(self); self->activities_.emplace_back(this); } } @@ -80,8 +45,8 @@ ExecImpl& ExecImpl::set_hosts(const std::vector& hosts) ExecImpl& ExecImpl::set_timeout(double timeout) { - if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) { - timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout)); + if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) { + timeout_detector_.reset(hosts_.front()->get_cpu()->sleep(timeout)); timeout_detector_->set_activity(this); } return *this; @@ -111,30 +76,41 @@ ExecImpl* ExecImpl::start() state_ = State::RUNNING; if (not MC_is_active() && not MC_record_replay_is_active()) { if (hosts_.size() == 1) { - surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front()); + surf_action_ = hosts_.front()->get_cpu()->execution_start(flops_amounts_.front(), bound_); surf_action_->set_sharing_penalty(sharing_penalty_); surf_action_->set_category(get_tracing_category()); - - if (bound_ > 0) - surf_action_->set_bound(bound_); } else { - surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1); + // get the model from first host since we have only 1 by now + auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model(); + surf_action_ = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1); } surf_action_->set_activity(this); + start_time_ = surf_action_->get_start_time(); } XBT_DEBUG("Create execute synchro %p: %s", this, get_cname()); return this; } +double ExecImpl::get_remaining() const +{ + if (state_ == State::WAITING || state_ == State::FAILED) + return flops_amounts_.front(); + return ActivityImpl::get_remaining(); +} + double ExecImpl::get_seq_remaining_ratio() { + if (state_ == State::WAITING) + return 1; return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost(); } double ExecImpl::get_par_remaining_ratio() { // parallel task: their remain is already between 0 and 1 + if (state_ == State::WAITING) + return 1; return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains(); } @@ -150,97 +126,102 @@ ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty) return *this; } +ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty) +{ + sharing_penalty_ = sharing_penalty; + surf_action_->set_sharing_penalty(sharing_penalty); + return *this; +} + void ExecImpl::post() { - if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */ - /* If the host running the synchro failed, notice it. This way, the asking + xbt_assert(surf_action_ != nullptr); + if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) { + /* If one of the hosts running the synchro failed, notice it. This way, the asking * process can be killed if it runs on that host itself */ state_ = State::FAILED; - } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) { - /* If the host running the synchro didn't fail, then the synchro was canceled */ + } else if (surf_action_->get_state() == resource::Action::State::FAILED) { + /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */ state_ = State::CANCELED; } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) { - state_ = State::TIMEOUT; + if (surf_action_->get_remains() > 0.0) { + surf_action_->set_state(resource::Action::State::FAILED); + state_ = State::TIMEOUT; + } else { + state_ = State::DONE; + } } else { state_ = State::DONE; } - get_iface()->set_finish_time(surf_action_->get_finish_time()); - clean_action(); timeout_detector_.reset(); - if (actor_) { - actor_->activities_.remove(this); - actor_ = nullptr; + if (get_actor() != nullptr) { + get_actor()->activities_.remove(this); } + if (state_ != State::FAILED && cb_id_ >= 0) + s4u::Host::on_state_change.disconnect(cb_id_); /* Answer all simcalls associated with the synchro */ finish(); } +void ExecImpl::set_exception(actor::ActorImpl* issuer) +{ + switch (state_) { + case State::FAILED: + static_cast(get_iface())->complete(s4u::Activity::State::FAILED); + if (issuer->get_host()->is_on()) + issuer->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed")); + else /* else, the actor will be killed with no possibility to survive */ + issuer->context_->set_wannadie(); + break; + + case State::CANCELED: + issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled")); + break; + + case State::TIMEOUT: + issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted")); + break; + + default: + xbt_assert(state_ == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s", + to_c_str(state_)); + } +} void ExecImpl::finish() { + XBT_DEBUG("ExecImpl::finish() in state %s", to_c_str(state_)); while (not simcalls_.empty()) { smx_simcall_t simcall = simcalls_.front(); simcalls_.pop_front(); + if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case + continue; // if process handling comm is killed + /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the * simcall */ + if (auto* observer = dynamic_cast(simcall->observer_)) { - if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case - continue; // if process handling comm is killed - if (simcall->call_ == simix::Simcall::EXECUTION_WAITANY_FOR) { - simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall); - size_t count = simcall_execution_waitany_for__get__count(simcall); - - for (size_t i = 0; i < count; i++) { - // Remove the first occurrence of simcall: - auto* exec = execs[i]; - auto j = boost::range::find(exec->simcalls_, simcall); - if (j != exec->simcalls_.end()) - exec->simcalls_.erase(j); - - if (simcall->timeout_cb_) { - simcall->timeout_cb_->remove(); - simcall->timeout_cb_ = nullptr; - } + const auto& execs = observer->get_execs(); + + for (auto* exec : execs) + exec->unregister_simcall(simcall); + + if (simcall->timeout_cb_) { + simcall->timeout_cb_->remove(); + simcall->timeout_cb_ = nullptr; } if (not MC_is_active() && not MC_record_replay_is_active()) { - ExecImpl** element = std::find(execs, execs + count, this); - int rank = (element != execs + count) ? element - execs : -1; - simcall_execution_waitany_for__set__result(simcall, rank); + auto element = std::find(execs.begin(), execs.end(), this); + int rank = element != execs.end() ? static_cast(std::distance(execs.begin(), element)) : -1; + observer->set_result(rank); } } - switch (state_) { - case State::DONE: - /* do nothing, synchro done */ - XBT_DEBUG("ExecImpl::finish(): execution successful"); - break; - - case State::FAILED: - XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname()); - simcall->issuer_->context_->set_wannadie(); - if (simcall->issuer_->get_host()->is_on()) - simcall->issuer_->exception_ = - std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed")); - /* else, the actor will be killed with no possibility to survive */ - break; - - case State::CANCELED: - XBT_DEBUG("ExecImpl::finish(): execution canceled"); - simcall->issuer_->exception_ = - std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled")); - break; - - case State::TIMEOUT: - XBT_DEBUG("ExecImpl::finish(): execution timeouted"); - simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted")); - break; - - default: - xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast(state_)); - } + + set_exception(simcall->issuer_); simcall->issuer_->waiting_synchro_ = nullptr; /* Fail the process if the host is down */ @@ -251,17 +232,23 @@ void ExecImpl::finish() } } +void ExecImpl::reset() +{ + hosts_.clear(); + bytes_amounts_.clear(); + flops_amounts_.clear(); + start_time_ = -1.0; +} + ActivityImpl* ExecImpl::migrate(s4u::Host* to) { if (not MC_is_active() && not MC_record_replay_is_active()) { resource::Action* old_action = this->surf_action_; - resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost()); + resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound()); new_action->set_remains(old_action->get_remains()); new_action->set_activity(this); new_action->set_sharing_penalty(old_action->get_sharing_penalty()); - - // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling. - // I hope that the user did not provide any. + new_action->set_user_bound(old_action->get_user_bound()); old_action->set_activity(nullptr); old_action->cancel(); @@ -273,6 +260,31 @@ ActivityImpl* ExecImpl::migrate(s4u::Host* to) return this; } +void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector& execs, double timeout) +{ + if (timeout < 0.0) { + issuer->simcall_.timeout_cb_ = nullptr; + } else { + issuer->simcall_.timeout_cb_ = timer::Timer::set(s4u::Engine::get_clock() + timeout, [issuer, &execs]() { + issuer->simcall_.timeout_cb_ = nullptr; + for (auto* exec : execs) + exec->unregister_simcall(&issuer->simcall_); + // default result (-1) is set in actor::ExecutionWaitanySimcall + issuer->simcall_answer(); + }); + } + + for (auto* exec : execs) { + /* associate this simcall to the the synchro */ + exec->simcalls_.push_back(&issuer->simcall_); + /* see if the synchro is already finished */ + if (exec->state_ != State::WAITING && exec->state_ != State::RUNNING) { + exec->finish(); + break; + } + } +} + /************* * Callbacks * *************/